Perplexity AI CEO Aravind Srinivas agrees that Computer Science is gradually returning to the domain of mathematics and physics

Aravind Srinivas, CEO of Perplexity AI, contributed to the growing discussion about AI's impact on software engineering. On March 13, he quote-tweeted a post by physics and AI/ML student @TheVixhal with the comment: “Well said.”

The original post, which garnered over 15,000 likes and nearly a million views, argued that large language models (LLMs) are quietly automating the repetitive aspects of coding. This shift is pulling computer science back toward its mathematical and physics-heavy roots.

This perspective is gaining traction. Anthropic CEO Dario Amodei has stated that the industry is 6 to 12 months away from AI handling most software engineering tasks end-to-end. He noted that some engineers at Anthropic no longer write code at all.

Replit’s CEO expressed this more directly, saying the software engineering role, as currently defined, “sort of disappears.”

GitHub Copilot boosts developers' speed by 55%—but gains are uneven

Data supports this shift. A 2023 Microsoft-run experiment found that developers using GitHub Copilot completed tasks 55.8% faster. Anthropic’s AI Exposure Index estimates that LLMs cover roughly 75% of programmers’ tasks—the highest coverage among professions tracked.

The change is not only about speed but also about what engineers focus on.

Routine boilerplate coding is increasingly handled by AI. Remaining challenges—such as understanding system failures, making trade-offs, and ensuring architecture scalability—are more aligned with physics and mathematics than with syntax.

LLMs still hallucinate on complex system design, keeping senior engineers in demand

However, not everyone believes this transition will be seamless. Critics highlight that LLMs still struggle with novel and complex problems. Junior developers benefit most from these tools, while senior engineers remain crucial for verification and critical judgment. The “6 to 12 months” timeline mentioned by Amodei applies to existing tasks, not the more difficult work of inventing new systems.

Nonetheless, the trend is clear enough that even Code.org’s founder is reconsidering computer science education—emphasizing less syntax and more logical reasoning. “Coding is dead,” he said, adding, “Long live coding.”